[QUOTE=ATH;411906]There is a real possibility a prime can be found during double check. It is not a big chance but much higher than I originally thought.

If you check this thread: [url]http://www.mersenneforum.org/showthread.php?t=20372[/url] Madpoo is finding lots of computers giving mostly bad results and most of the results are proved wrong when people are double/triple/quad checking them.[/QUOTE]

Yeah, so far my overall success* rate has been about 33%. I had a string of even better results (60%+) the other day which was fun. I think others have had even better success* since I've been trying to leave the easier pickings for other people.

* = success meaning I got a different (correct?) residue than the first check

[QUOTE=ATH;411906]There is a real possibility a prime can be found during double check. It is not a big chance but much higher than I originally thought.

If you check this thread: [url]http://www.mersenneforum.org/showthread.php?t=20372[/url] Madpoo is finding lots of computers giving mostly bad results and most of the results are proved wrong when people are double/triple/quad checking them.[/QUOTE]

Thank you.
Is the possibility of not finding a prime after 2 LL tests, comparable to that of not finding a prime in the first LL test (within a range lets say exponent 60,000,000 to 80,000,000)

In other words how sure are we after doing the first LL run that there are no primes in the range? 98%? 99%? 99.9%? 99.999?

I hope the question makes any sense.:smile:

[QUOTE=rudy235;411937]Thank you.
Is the possibility of not finding a prime after 2 LL tests, comparable to that of not finding a prime in the first LL test (within a range lets say exponent 60,000,000 to 80,000,000)

In other words how sure are we after doing the first LL run that there are no primes in the range? 98%? 99%? 99.9%? 99.999?

I hope the question makes any sense.:smile:[/QUOTE]

Basically what you'd want to know is: What is the average error rate for first time checks?

The answer is a little murky, and that's because we don't know the error rate until we've swept through and done the double-checking.

We can say for sure that the error rate for exponents below 34,661,527 (everything below that has been DC'd) is:
61,047 bad out of a total 746,062 exponents tested at least twice (at least 1,492,124 tests) = ~4.1 % error rate

In truth, those 746,062 exponents have 1,574,499 tests done... some were done more than twice, including the triple checks by yours truly and others, or people just doing quality control tests.

But if we break it down into smaller chunks, not just the grand total from 2 up to the latest DC, but let's say maybe 30M up to 34.6M:
7886 bad out of 196855 good tests = 4.0 %... slightly lower.

Similarly between 20e6 and 30e6: 17,307 bad out of 438,199 good tests = 3.95% failures.

Now, if we look at the relatively small # of exponents above 40M that have been DC'd so far, we get:
Bad = 599, Good tests (total) = 21,290 for an error rate of "only" 2.8%. But as you might guess, that's misleading because there are still so many untested first time checks out there, we don't really know the total error rate. Based in historical averages, it's probably still going to be around 4%, we just won't know until many years from now.

Anyway, the #'s do change over time as we push hardware to new levels of exertion (larger FFTs using more memory, or maybe when GPUs got into the LL game). But it still seems to hover in the 3.9%-4.1% range no matter which range of exponents I look at.

And honestly, it would be beyond what I feel like doing to analyze whether the bad check in those cases was the first one or if it was done later... in many cases for those smaller exponents we don't have the exact date it was checked in anyway, so it would be hard to tell. If I had to guess, and based on my work with doing needed triple-checks of exponents in the 34M range (I've done a few hundred now), it seems like it's half and half. In about half of the cases, the first test was bad, and the other half the double-check that was done by someone was bad and my triple-check proved the first one was correct.

That's anecdotal and based on my spotty memory of what I saw when looking at results for a few seconds each. But if I'm even close, that means maybe 2% were done wrong the first time.

Which makes it even more impressive that I've been able to crunch some stats on machines and do strategic double-checking with (for me) a 30-40% rate of finding those bad machines. And hopefully the others who have been helping out have had even better rates of success with that, over 50% if I'm crunching the numbers right... :smile:

I even did some queries to look at exponents that have been double-checked with a mismatch, and where *both* systems are in the "suspect" category. But honestly, I only found a few of them and the query is so "expensive" (lots of CPU) that I dare not run it too often. LOL But it has given me a few where my triple-check matched neither, so I've mentioned the exponent if anyone wanted to do the quad check. :smile:

[QUOTE=Madpoo;411942]
Similarly between 20e6 and 30e6: 17,307 bad out of 438,199 good tests = 3.95% failures.

Now, if we look at the relatively small # of exponents above 40M that have been DC'd so far, we get:
Bad = 599, Good tests (total) = 21,290 for an error rate of "only" 2.8%. [/QUOTE]
Very nice statistics, thanks. Note that this comes in line with what George said in the very beginning in the math page of GIMPS server, that the historical error rate is around 3%. Right now, you support that very well, with numbers.

Ignoring exponents with duplicate tests (a third confirmation or multiple bad tests), what's the ratio of exponents with any nonzero number of bad tests to total exponents?

[QUOTE] Countdown to testing all exponents below M(57885161) once: 0
Countdown to first time checking all exponents below 58M: 0 ()[/QUOTE]
:party:

[B]2015-10-04 All exponents below M(57885161) tested at least once. 14 discovered 14 tested[/B]
2013-01-25 Prime M(57885161) discovered!! 14 discovered 13 tested
2012-09-05 All exponents below M(43112609) tested at least once. 13 discovered 13 tested
2012-09-05 All exponents below M(42643801) tested at least once. 13 discovered 12 tested
2010-12-25 All exponents below M(37156667) tested at least once. 13 discovered 11 tested
2010-12-25 All exponents below M(32582657) tested at least once. 13 discovered 10 tested

[QUOTE=Madpoo;411694]I poached one of them. It hadn't reported in for about a month, hadn't started yet at the time. It seemed abandoned by user "Xebecer"? Well, that assignment is now a DC.[/QUOTE]

Not anymore. Someone apparently with a valid assignment ID from 6 months ago has completed the double-check. [url=http://www.mersenne.org/report_exponent/?exp_lo=57398083&exp_hi=&full=1]M57398083[/url]

[QUOTE=cuBerBruce;411960]Countdown to testing all exponents below M(57885161) once: 0
Countdown to first time checking all exponents below 58M: 0 ()[/QUOTE]

And I just updated the milestone page with the latest dates.

While I was at it, I swapped over to the cached version of the milestone info and added in the 60M and 61M countdowns so we have more to obsess over. :smile:

[QUOTE=cuBerBruce;411971]Not anymore. Someone apparently with a valid assignment ID from 6 months ago has completed the double-check. [url=http://www.mersenne.org/report_exponent/?exp_lo=57398083&exp_hi=&full=1]M57398083[/url][/QUOTE]

That's definitely weird. Oh well... at least Xebecer still doesn't seem to have checked anything in, which I assumed would be the case, but what odd timing that someone with an expired assignment on that *finally* after months and months checks something in at long last, shortly after I did. Weird.

[QUOTE=Madpoo;411988]And I just updated the milestone page with the latest dates.[/QUOTE]This line still appears in the report:[code]• Countdown to testing all exponents below M(57885161) once:
[/code]